Let $M $ be a matrix in $ \operatorname{GL}(2, \mathbb{Z})$ that has at least one eigenvalue of absolute value strictly bigger than $1$. What are the finite index subgroups $H$ of $\mathbb{Z}^2$ such that for all $v \in H$, we have $M(v) \in H$?

Question: One example of such subgroup has the form $H=\langle (a,0),(0,a)\rangle$, for a non-zero integer $a$, this example works for any matrix in $ \operatorname{GL}(2, \mathbb{Z})$. I was wondering if there are any other such subgroups.

Thought so far: I think the finite index subgroups of $\mathbb{Z}^2$ have the form $H = \langle (a,b),(c,d)\rangle$, where $a,b,c,d \in \mathbb{Z}$ such that $ad-bc\neq 0$. Hence, in order for $H$ to be a subgroup, we need $M(a,b)= \alpha_1(a,b) + \alpha_2(c,d)$ and $M(c,d)= \beta_1(a,b) + \beta_2(c,d)$ for some integers $\alpha_i$ and $\beta_i$, but I can't see any obvious solutions for these equations.

I was also thinking maybe we can use the eigenvectors, but unfortunately the eigenvectors might not have integers entries.

Any ideas for constructing such a subgroup would be really appreciated.

  • 2
    $\begingroup$ Such a subgroup contains $n\mathbf{Z}^2$. So the question "reduces" to an understanding of determining, for all $n$, invariant subgroups of $(\mathbf{Z}/n\mathbf{Z})^2$ under the reduction of $M$ modulo $n$. So the answer should be mostly of arithmetic nature. $\endgroup$
    – YCor
    Commented May 7, 2022 at 18:21
  • 1
    $\begingroup$ The order of the quantifiers is not clear to me. Are you asking, for a given $M$, which subspaces $H$ are stable under it or, for a given $H$, whether it is stable under some $M$? $\endgroup$
    – LSpice
    Commented May 7, 2022 at 19:43
  • 1
    $\begingroup$ @LSpice sorry, I meant to ask "for a given $M$, which subgroups $H$ are invariant under it" $\endgroup$
    – ghc1997
    Commented May 7, 2022 at 19:55

3 Answers 3


If $v$ is not scalar, the ring $\mathbb Z[M]$ generated by $M$ has rank $2$ over $\mathbb Z$. It is therefore either an order in a number field, a finite-index subring of $\mathbb Z^2$, or $\mathbb Z[\epsilon]/\epsilon^2$.

$\mathbb Z^2$ is then a module over $\mathbb Z[M]$, and you are looking to describe the finite-index submodules of $\mathbb Z[M]$.

In the simplest case, $\mathbb Z^2$ is a (locally) free module over $\mathbb Z[M]$, and so you are describing the ideals of $\mathbb Z[M]$. In the ring of integers in a number field case, this has a particularly simple description, as products of prime ideals.

In the general order case, submodules are given by products of prime ideals away from those primes where the order $\mathbb Z[M]$ is not maximal or the module $\mathbb Z^2$ is not locally free, and there is additional complexity at those primes. Said another way, the simplest construction of such a subgroup is to take the product of $\mathbb Z^2$ with some ideal of the order $\mathbb Z[M]$, for example the intersection of $\mathbb Z[M]$ with an ideal of the ring of integers of the number field.

In the other cases there is a similarly concrete description.

Edit: Actually maybe I should say this a different way. Your idea of using the eigenvectors is a good one, but it's better to use the eigenvectors modulo $p$. Consider the characteristic polynomial of the element $M$, and its discriminant. If this discriminant is nonzero then there are infinitely many primes $p$ modulo which the discriminant is a nonzero square (quadratic reciprocity + Dirichlet's theorem). When the entries of $M$ are taken mod $p$, allowing us to view $M$ as a matrix over $\mathbb F_p$, it has two distinct eigenvalues and therefore two eigenvectors. The subgroup of vectors congruent mod $p$ to a multiple of one of these two eigenvectors is an index $p$ subgroup invariant under $M$.

If the discriminant is $0$ and $M$ is non-scalar, so $M$ has a single $2 \times 2$ Jordan block, then modulo all but finitely many primes $p$, $M$ will still have a single $2\times 2$ Jordan block, and you can take elements congruent mod $p$ to multiples of the unique eigenvector. Modulo the other primes, where $M$ becomes scalar, you can take multiples of any vector.


To illustrate how such a question is of arithmetic nature (and what a complete answer should look like), here is a partial answer in a specific case.

Namely: I specify to $M=\begin{pmatrix}2 & 1 \\ 1 & 1\end{pmatrix}$ and I address the question: what are invariant finite index subgroups of prime index.

The characteristic polynomial of this matrix is $X^2-3X+1$, which has discriminant 5. By the quadratic reciprocity formula, for an odd prime $p\neq 5$, 5 is a square mod $p$ iff $p=\pm 1$ mod $5$ (i.e. the decimal expansion of of $p$ terminates with $1$ or $9$). And this polynomial is irreducible mod $2$, and has a double root mod $5$.

Hence, for a prime $p$:

  • for $p=\pm 2$ mod $5$, there is no $M$-invariant subgroup of index $p$ in $\mathbf{Z}^2$;
  • for $p=\pm 1$ mod $5$, there are exactly two $M$-invariant subgroups of index $p$ in $\mathbf{Z}^2$;
  • there is a single $M$-invariant subgroup of index $5$ in $\mathbf{Z}^2$.


For a general matrix $M$, the characteristic polynomial is $X^2+nX\pm 1$ for some $n$ and there should be a similar discussion.

For a general index $q$, one should boil down to when $q$ is a power of a prime $p$. Then in turn one should boil down to when the quotient is cyclic (the quotient being isomorphic to $C_{p^a}\times C_{p^b}$ for some $a\le b$, the subgroup is contained in $p^a\mathbf{Z}^2$ and we can then "replace" $\mathbf{Z}^2$ with $p^a\mathbf{Z}^2$ to assume $a=1$).


For A fixed $M$, such finite index subgroups may be categorized as follows. Let $W\in GL(2,\mathbb{Z})$ and $b$ a positive integer dividing the lower left element of the matrix $W^{-1} M W$. Then the columns of the matrix $W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}$ generate a subgroup $H$ of $\mathbb{Z}^2$ (of finite index $b$) for which $Mv\in H$ for any $v\in H$. Conversely, all such subgroups for this $M$ arise this way, up to scaling the subgroup by an integer $a$ as already noted in the question.

To see this, first note that a finite index subgroup of $\mathbb{Z}^2$ may be generated by the columns of a non-singular integer matrix $A$, known as a basis. Any other basis equals $AU$, where $U$ is unimodular, i.e. an element of $GL(2,\mathbb{Z})$.

For unimodular $M$, you want to know for which such $A$ do we have $MA = AV$, for some integer matrix $V$, as such $V$ just makes integer linear combinations of the columns of the basis $A$, which give elements of the finite index subgroup $H$. But determinants then imply $V$ is also unimodular, as it has the same determinant as $M$. So $M$ takes any basis of $H$ to another basis of $H$.

Integer matrices have a Smith Normal Form: there exist unimodular $W,V$ so that $W\begin{bmatrix} a & 0\\0 & ab\end{bmatrix}V = A$, where $a,b$ are integers (here positive, as the determinant is non-zero). As any choice of basis will do, we choose the basis $W\begin{bmatrix}a & 0\\0 & ab\end{bmatrix}$.

So an $H$ may be specified by a unimodular $W$ and integers $a$ and $b$. As $a$ simply scales everything we henceforth assume WLOG that $a=1$.

We now characterize such pairs $W, b$. We have from before the equation $$ M (W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}) = (W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}) V$$ so $$ (W^{-1} M W) \begin{bmatrix} 1 & 0\\0 & b\end{bmatrix} = \begin{bmatrix} 1 & 0\\0 & b\end{bmatrix} V$$

As $W$ is unimodular, it has integer inverse, so all matrices above are integer. The matrix on the left has second column divisible by $b$, while the matrix on the right has second row divisible by $b$. As they equate, both sides must equal a matrix of the form $\begin{bmatrix} x & by\\bz & bw\end{bmatrix}$ where $w,x,y$ and $z$ are integers. But then $V$ equals $\begin{bmatrix} x & by\\z & w\end{bmatrix}$, so $V$ unimodular implies $xw - byz = \pm 1$.

Similarly, we have $W^{-1} M W = \begin{bmatrix} x & y\\bz & w\end{bmatrix}$, since the diagonal matrix on the right simply scales the second column by $b$. So we started with a general subgroup $H$ invariant under $M$ - generated by the columns of a matrix $A$ - and showed that this $H$ has a basis of the form $W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}$ where the index $b$ divides the bottom left entry of $W^{-1} M W$, as claimed.

For the other direction, let $b$ be a positive factor of the bottom left entry of $W^{-1} M W$, where $W$ is any unimodular matrix. This means

$$W^{-1} M W = \begin{bmatrix} x & y\\bz & w\end{bmatrix}$$

for some integers $w, x, y$ and $z$, where $xw - bzy =\pm 1$, as the matrix is unimodular. But then $$ M W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix} = W \begin{bmatrix} x & y\\bz & w\end{bmatrix}\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix} $$ and $$M \left(W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}\right)= W\begin{bmatrix} x & by\\bz & bw\end{bmatrix}=W \begin{bmatrix} 1 & 0\\0 & b\end{bmatrix} \begin{bmatrix} x & by\\z & w\end{bmatrix}= \left(W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}\right) V $$

where the matrix $V$ is unimodular, as we have already noted $xw - bzy =\pm 1$. But this means the subgroup $H$ of $\mathbb{Z}^2$ generated by the columns of $ W\begin{bmatrix} 1 & 0\\0 & b\end{bmatrix}$ is invariant under $M$, as claimed.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.